5.3% of the Vessel Graph Is Mutation Edges. Shell Hops Average 13-Day Tenure. That Asymmetry Is the Screening Problem.

The Setup

The vessel temporal graph currently holds 25,352 edges across eleven relation types. Most screening workflows treat these edges as interchangeable — a SHIP_MANAGER link gets the same analytical weight as a shell_hop. That is a calibration error baked into the graph's own shape.

Three relation types — SHIP_MANAGER (32.1%), OWNED_BY (32.1%), and flagged_as (30.5%) — account for 94.7% of all edges. The remaining active mutation types (renamed_to at 2.2%, mmsi_changed at 2.0%, shell_hop at 1.1%) compose the 5.3% that carries the identity obfuscation signal.

The Chain

The tenure differential makes the asymmetry concrete. Stable backbone edges — SHIP_MANAGER, OWNED_BY, flagged_as — average 27 days of tenure, consistent with the catalog's ingest window. Mutation edges are significantly shorter: renamed_to sits at 17 days, mmsi_changed at 14 days, shell_hop at 13 days.

That compression is not coincidental. Shell hops involve structural ownership chain restructuring — new entities introduced, old links severed — and the graph captures them at the moment of transition. The 13-day average tenure for shell_hop edges reflects how quickly these transitions cycle through the catalog before being superseded by new ownership records.

The 269 shell_hop edges in the current graph represent distinct ownership chain restructuring events. Compared to the 515 MMSI changes and 546 renames, shell hops are the rarest mutation type but carry the highest structural disruption per event — each hop requires coordinated changes across at minimum OWNED_BY and typically SHIP_MANAGER as well. A single shell hop event touches multiple edge types simultaneously.

The 5.3% mutation edge share is not a small rounding error. At 1,330 edges, it is a meaningful catalog layer — one that has grown from the 1,005 mutation events counted 10 days ago, suggesting active accumulation in the graph's current ingest window.

The Implication

Screening models that score vessels by aggregating edge counts are running against the wrong prior. The graph is 94.7% structural backbone. An algorithm weighting by raw edge volume will allocate 94.7% of its attention to stable relationships and 5.3% to the edge types where adverse events concentrate.

The correct calibration inverts that priority. Mutation edge presence — particularly shell_hop — should act as a multiplier on a vessel's screening priority, not a marginal increment in an additive edge count. A vessel with a single shell_hop edge in the last 13 days warrants higher immediate priority than a vessel with 40 stable OWNED_BY entries and no mutation history.

This is the directional claim: the vessel graph's topology is adversarially shaped. The 5.3% mutation edge layer exists precisely because it is difficult to detect within a graph that is 94.7% background noise.

What to Watch

As the temporal graph extends its ingest window beyond the current ~27-day horizon, the tenure differential between stable and mutation edges will widen. Long-running SHIP_MANAGER relationships will accumulate tenure while shell_hop events will continue cycling at 13-day average. The spread between edge-type tenures is a live signal: widening spread indicates more rapid restructuring in the mutation layer relative to a stable backbone.

The ratio of shell_hops to mmsi_changed is also worth tracking. Currently 269:515 (roughly 1:2), this ratio reflects how many identity restructurings require full ownership chain reorganization versus a simpler identifier swap. A rising shell_hop-to-MMSI ratio would indicate more sophisticated obfuscation activity.

Limitations

Edge classifications derive from commercial registry feeds with a known lag of up to 14 days for ownership transitions. Short-tenure shell_hop events completed within the lag window may not appear in the graph until the next feed update. The 27-day average tenure for stable edges reflects the current ingest window, not long-term relationship duration — extending the window will revise this baseline upward and make the stable-vs-mutation tenure gap larger. Mutation edge counts reflect unique graph edges, not unique vessels; a single vessel can contribute multiple mutation edges across a single restructuring event.

Graph snapshot: 2026-05-15. 25,352 total edges, 11 relation types. Stable backbone (SHIP_MANAGER, OWNED_BY, flagged_as): 24,017 edges. Mutation layer (shell_hop, mmsi_changed, renamed_to): 1,330 edges.